Photoacoustic Microscopy (PAM) offers a unique combination of high contrast from endogenous optical absorbers, and increased penetration to image microvasculature. However, images of the vasculature at increased depth are often corrupted by acoustic reverberation from superficial layers. In this paper, we present an algorithm using dictionary learning to remove the reverberant signal while preserving underlying microvascular anatomy. The algorithm was validated in vitro, using dyed beads embedded polydimenthylsiloxane (PDMS). Subsequently, we demonstrate suppression of reverberant artifacts by 20 ± 0.2 dB using in vivo PAM data acquired in a mouse brain.